ECP+TR

class globe.optimizers.decision.ECP.ECP(bounds, n_eval=50, epsilon=0.01, theta_init=1.001, C=1000, max_trials=10000000, trust_region_dict={'bobyqa_eval': 20, 'radius': 0.1}, verbose=False)[source]

Bases: Decision_Optimizer

Interface for the ECP+TR optimizer.

Parameters:

  • bounds (ndarray) – The bounds of the search space.

  • n_eval (int) – The maximum number of function evaluations.

  • epsilon (float) – The initial Lipschitz constant estimate.

  • theta_init (float) – The scaling factor for epsilon.

  • C (float) – How many candidates to sample before increasing epsilon.

  • max_trials (int) – The maximum number of potential candidates sampled at each iteration.

  • trust_region_dict (dict, optional) –

    Dictionary with the following keys:

    radiusfloat

    The trust region radius.

    bobyqa_evalint

    The number of evaluations for the BOBYQA optimizer.

    If None, the trust region is not used.

  • verbose (bool) – Whether to print information about the optimization